[
https://issues.apache.org/jira/browse/HIVE-7950?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14127901#comment-14127901
]
Josh Elser commented on HIVE-7950:
----------------------------------
I think I finally got to the bottom of this, and it is broken with Tez-0.5.0.
TezTask needs to be altered (as described by the previous discussion) to add
the necessary StorageHandler resources to the DAG using
{code}
dag.addTaskLocalFiles(localResources);
{code}
For the case of the AccumuloStorageHandler, this adds the jars necessary to
connect to Accumulo to {{commonTaskLocalFiles}} in {{DAG}}. Then, {{TezTask}}
will proceed to eventually submit the DAG to be run.
{code}
try {
// ready to start execution on the cluster
sessionState.getSession().addAppMasterLocalFiles(resourceMap);
dagClient = sessionState.getSession().submitDAG(dag);
} catch (SessionNotRunning nr) {
console.printInfo("Tez session was closed. Reopening...");
// close the old one, but keep the tmp files around
TezSessionPoolManager.getInstance().closeAndOpen(sessionState, this.conf);
console.printInfo("Session re-established.");
dagClient = sessionState.getSession().submitDAG(dag);
}
{code}
Consider the case where we had a Session already created for the user, but the
underlying application has exited, say due to a timeout. In the try block, we
try to submit our DAG to run. In doing so, TezClient creates a DAGPlan from the
DAG
{code}
DAGPlan dagPlan = dag.createDag(amConfig.getTezConfiguration());
{code}
When we create a {{DAGPlan}} from the {{DAG}}, we modify the {{DAG}} instance,
adding the local resources to each {{Vertex}} in the {{DAG}}. Then, we identify
that the underlying application has already died, and that we need to
{{closeAndOpen}} a new Session. So, we get the {{SessionNotRunning}} exception,
pop out to the catch block, and end up creating another {{DAGPlan}} from the
{{DAG}} _that was already altered by the last attempt to submit it_.
As I'm looking at it, I don't think there's anything I can do at the Hive level
to fix this because {{TezClient}} will always try to add duplicate resources to
the {{Vertex}}'s in a {{DAG}} which throws an Exception and tanks the query.
> StorageHandler resources aren't added to Tez Session if already Session is
> already Open
> ---------------------------------------------------------------------------------------
>
> Key: HIVE-7950
> URL: https://issues.apache.org/jira/browse/HIVE-7950
> Project: Hive
> Issue Type: Bug
> Components: StorageHandler, Tez
> Reporter: Josh Elser
> Assignee: Josh Elser
> Fix For: 0.14.0
>
> Attachments: HIVE-7950-1.diff, hive-7950-tez-WIP.diff
>
>
> Was trying to run some queries using the AccumuloStorageHandler when using
> the Tez execution engine. Some things that classes which were added to
> tmpjars weren't making it into the container. When a Tez Session is already
> open, as is the normal case when simply using the `hive` command, the
> resources aren't added.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
